Scholarlink

In this paper we present an improved Content-Based Image Retrieval (CBIR) system based on deep visual features, statistical Feature Re-Weighting, as well as iterative Relevance Feedback to improve retrieval performance. Pretrained ResNetRS-50 model based features are computed and are Re-Weighted by Mutual Information among them. First, search is conducted by cosine similarity and an iterative Relevance Feedback cycle inspired by Rocchio algorithm is employed to refine a query vector. Experiments on the Oxford 17 Flowers dataset show significant improvements on Precision@10 for iterations with respect to the mAP score of 0.91, outperforming previous best results in 8%. Such results highlight the system’s potential to narrow the Semantic Gap, providing a scalable solution for fine-grained image retrieval problems.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Semantic Gap Reduction Using Relevance Feedback and Feature Reweighting for Content Based Image Retrieval Using ResNetRS

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Varkala Satheesh Kumar,
  • Vijaya Chandra Jadala

摘要

In this paper we present an improved Content-Based Image Retrieval (CBIR) system based on deep visual features, statistical Feature Re-Weighting, as well as iterative Relevance Feedback to improve retrieval performance. Pretrained ResNetRS-50 model based features are computed and are Re-Weighted by Mutual Information among them. First, search is conducted by cosine similarity and an iterative Relevance Feedback cycle inspired by Rocchio algorithm is employed to refine a query vector. Experiments on the Oxford 17 Flowers dataset show significant improvements on Precision@10 for iterations with respect to the mAP score of 0.91, outperforming previous best results in 8%. Such results highlight the system’s potential to narrow the Semantic Gap, providing a scalable solution for fine-grained image retrieval problems.